65% of Employees Want to Roll Back Workplace AI. The Tech Isn’t the Real Problem.

Well, this is awkward.

After several years of companies racing to buy AI tools, announce AI initiatives, create AI task forces, roll out copilots and encourage everyone to “experiment with AI,” a new survey of 2,500 knowledge workers found that 65% regularly feel nostalgic for how work operated before widespread AI adoption.

Here’s the interesting part: 67% also want their organizations to increase the use of AI.

That sounds contradictory. I don’t think it is.

Employees don’t hate AI. They hate bad AI transformation.

And frankly, who can blame them?

For a lot of employees, the corporate AI adoption experience has gone something like this:

Here is your new AI tool. Please use it. Here is a 60-minute webinar. Remember not to put anything confidential into it. Also, we expect productivity to increase. Good luck!

That’s not transformation.

That’s software distribution.

We Bought the AI. Why Isn’t Everyone Transformed?

This is the part of the AI conversation I think we’ve gotten backwards.

Organizations have spent enormous amounts of time deciding which technology to buy.

ChatGPT Enterprise or Copilot? Claude or Gemini? Which agents? Which platform? Which model? Which license?

Those are important decisions. But eventually somebody has to answer a much less exciting question: How exactly are we going to work differently on Monday morning?

That’s where things get harder. Because AI doesn’t magically redesign a workflow when you purchase a license.

It doesn’t decide which parts of someone’s job should remain human. It doesn’t determine when AI output requires review.

It doesn’t resolve the fact that Susan has developed an elaborate Claude workflow, Michael is doing essentially the same thing in Copilot, and Denise has created 37 prompts that nobody else knows exist.

And it definitely doesn’t walk into the weekly team meeting and say: “You know, half the steps in this process don’t make sense anymore.”

Someone has to redesign the work.

The Problem Isn’t Adoption. It’s Integration.

Prosci recently reported something particularly interesting.

Employee motivation to use AI was nearly the same in organizations succeeding with AI and organizations struggling with it.

Think about that.

The difference wasn’t simply having employees who were more excited about AI.

The successful organizations were better at creating the conditions that turned individual AI adoption into organizational value.

And that tracks closely with something we’ve been seeing for three years.

Employees Didn’t Wait for the Organization

Over the past three years, Human Driven AI has studied how employees actually adopt generative AI inside their day-to-day work.

One pattern has become impossible to ignore: Employees moved faster than their organizations did.

They didn’t wait for an enterprise AI strategy. They didn’t wait for the governance committee. They didn’t wait for someone to redesign their workflows. They opened ChatGPT, Copilot, Claude or another tool and started figuring out what worked.

That’s good. It created experimentation, learning and some genuinely impressive productivity gains.

But it also created something organizations are only now beginning to reckon with.

If you have 59 employees experimenting independently with AI, you may not have one AI transformation underway. You may have 59 different AI workflows.

One employee has figured out how to conduct research in 20 minutes instead of two hours. Another has built a fantastic prompt sequence for drafting client reports. Someone else has created a Claude Project. Another employee has built a custom GPT.

And somewhere in accounting, Gary may be doing something with AI that absolutely nobody knows about. Gary might be a genius. Gary might also be uploading the quarterly financials. We should probably talk to Gary.

The point isn’t that decentralized experimentation was a mistake. In many cases, it was exactly how organizations began discovering AI’s value.

The problem is what happens when companies leave AI adoption at the individual level.

The organization doesn’t capture what employees have learned. Successful workflows aren’t standardized or shared. Duplicate efforts multiply. Institutional knowledge remains trapped inside individual accounts and chat histories. And governance becomes something employees are expected to interpret individually rather than something embedded into the way work gets done.

That’s why I don’t think the next phase of enterprise AI is primarily an adoption challenge.

It’s a convergence challenge.

Organizations now have to catch up to their own employees.

They need to find the AI practices already happening across the business, determine which ones are actually valuable, redesign them where necessary, establish shared standards and turn the best individual practices into organizational capabilities.

Because 59 employees creating 59 workflows can be an incredibly valuable discovery phase. But it shouldn’t be the operating model. Companies have spent the last few years asking: How do we get more employees to use AI?

The better question now is: How do we turn what our employees have already learned into a better way of working together?

The first produces individual adoption.

The second produces transformation.

Your Best AI Users May Be Creating Another Problem

There’s another wrinkle here.

In many organizations, the people who enthusiastically adopted AI have spent the last two or three years creating their own systems. Their own prompts. Their own GPTs. Their own Claude Projects. Their own Copilot workflows. Their own shortcuts. Their own weird little AI ecosystems.

Some of them are brilliant. But very few are shared.

So instead of eliminating organizational silos, AI can quietly create a brand-new generation of them.

Now imagine an employee leaves. Does the workflow leave too? Does anyone know how she produced that weekly analysis in 20 minutes? Does anyone know which sources were feeding it? Does anyone know what prompts she used? Does anyone know where human review happened?

Sometimes the answer is essentially: “No, but we think it’s somewhere in her ChatGPT history.”

That isn’t institutional knowledge.

That’s a scavenger hunt.

This Gets Even More Important With AI Agents

And now we’re entering the agentic AI era.

Marketing organizations aren’t just giving employees AI assistants anymore. They’re beginning to deploy AI systems capable of completing multiple steps of work.

Which means the consequences of poorly designed work become much bigger.

An employee using AI badly might produce a mediocre draft. An agent operating inside a poorly designed process can produce mediocre work at scale and with impressive efficiency.

Congratulations. We automated the problem.

Before organizations start dropping agents into every available workflow, they need to understand the workflow itself.

• What triggers the work?
• What information does AI need?
• Which systems can it access?
• Which decisions can AI make?
• Which decisions require human judgment?
• Where does approval happen?
• What gets documented?
• What happens when something goes wrong?

And perhaps most importantly: Which parts of the existing process shouldn’t exist anymore?

Because automating a bad workflow doesn’t make it a good workflow.

It just makes the bad workflow faster.

Training Still Matters. But Training Has to Change.

This doesn’t mean organizations should stop AI training.

Quite the opposite.

But we’re rapidly moving past the era when useful enterprise AI training can consist primarily of: Here’s what generative AI is. Here’s how to write a prompt. Now everyone try one.

Employees need training tied to their actual work.

• A communications team should be working through media monitoring, messaging, executive communications, content development and crisis workflows.
• A marketing team should be looking at research, campaign development, content production, analytics and customer insights.
• A fundraising team should be working through donor research, portfolio management and stewardship.

The point isn’t simply to teach people how to operate AI.

It’s to help them understand where AI belongs in their work, and where it doesn’t.

And increasingly, training shouldn’t happen separately from workflow design.

If we’ve determined how a team wants research, content development or reporting to work with AI, training should help employees practice that shared workflow, not send everyone back to their desks to invent another one.

Governance Can’t Live in a PDF Either

The same problem exists with AI governance.

Organizations write an AI policy. They publish it. Everyone clicks the button confirming they read it. And technically, governance has occurred.

Except employees still have to make dozens of judgment calls every week.

• Can I upload this?
• Can AI summarize that?
• Can I use this customer information?
• Do I need to disclose AI involvement?
• Who reviews this output?
• Where should this be stored?
• What happens when the AI is wrong?

Governance becomes real when those decisions are embedded in workflows, not when the policy is sitting somewhere on SharePoint.

A policy can establish the rules.

The workflow is where employees actually have to apply them.

The Next Phase of AI Transformation Is About Work

The companies that win the next phase of AI aren’t necessarily going to be the companies with the most AI tools. Or the most agents. Or the largest number of employees using ChatGPT.

They’re going to be the organizations that figure out how humans and AI should work together.

That requires connecting four things that have too often been treated separately: Governance. Workflows. Training. Automation.

• Governance establishes the boundaries.
• Workflow redesign determines how the work should happen.
• Training gives people the skills and judgment to operate within the new system.
• Automation scales the parts that actually should be automated.

And the order matters.

Because starting with automation and figuring out the rest later is how you end up with employees asking to roll AI back.

After three years of rapid experimentation, employees have already shown organizations where AI can be useful.

Now organizations have a different job: Capture what they’ve learned. Make decisions about how the work should change. Build governance into those workflows. Train teams around shared practices. Then automate what deserves to be automated.

The goal was never supposed to be more AI.

The goal was supposed to be better work.

Maybe it’s time we started designing for that.

---

Remember: Buying technology is easy. Changing how people work is harder.

Human Driven AI helps organizations move from scattered AI experimentation to shared, scalable practice. We build the governance foundations, redesign workflows around human and AI strengths, and deliver custom training programs that teach your teams how to put those systems into practice.

Whether you need an executive Lunch-and-Learn, a hands-on offsite workshop, an enterprise AI transformation program or a strategy to strengthen your brand’s visibility through GEO, we help you turn AI capability into better ways of working.

Ready to move from using AI to working differently with it? Let’s talk.

Read more: 65% of Employees Want to Roll Back Workplace AI. The Tech Isn’t the Real Problem.